GPU Rendering - NVIDIA GeForce RTX 30 Series Multi-GPU Scaling

Always look at the date when you read an article. Some of the content in this article is most likely out of date, as it was written on November 16, 2020. For newer information, see our more recent articles.

Table of Contents

TL;DR: NVIDIA GeForce RTX 30 Series Multi-GPU Rendering Performance

Running multiple of NVIDIA's GeForce RTX 30 Series video cards provides excellent scaling in GPU-based rendering engines!

This is no surprise, as it was true with the RTX 20 Series and other GeForce and Quadro video cards as well, but it is still impressive to see what these cards are capable of when used in sets of two or three. However, it is important to keep cooling and power consumption in mind when putting together a multi-GPU workstation. Most graphics cards do not have cooling systems that are built to be used in tandem with others, and the RTX 30 Series draw more power under load than their predecessors.

Introduction

With the initial launches in NVIDIA's GeForce RTX 30 Series complete, and availability of the RTX 3070, 3080, and 3090 getting better, it is time to look at how well these cards scale in multi-GPU configurations for rendering within Redshift, OctaneRender, and V-Ray.

Three Gigabyte NVIDIA GeForce RTX 3090 Turbo video cards on an open test bed in Puget Labs

Three Gigabyte NVIDIA GeForce RTX 3090 Turbo video cards on an open test bed in our lab

Rendering Workstations

Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Test Setup

Listed below are the specifications of the system we used for our GPU scaling tests:

Test Platform
CPU	AMD TR 3960X 24 Core
CPU Cooler	Noctua NH-U14S TR4-SP3
Motherboard	Gigabyte TRX40 AORUS PRO WIFI
RAM	4x DDR4-3200 16GB (64GB total)
Video Card	GIGABYTE NVIDIA GeForce RTX 3090 TURBO 24GB NVIDIA GeForce RTX 3080 Founders Edition 10GB NVIDIA GeForce RTX 3070 Founders Edition 8GB
Hard Drive	Samsung 960 Pro 1TB
Software	Windows 10 Pro 64-bit V-Ray Benchmark 1.0.8 V-Ray Next Benchmark 4.10.06 OctaneBench 2020.1.5 Redshift Demo 3.0.28 NVIDIA GeForce Driver 457.30

You may have noticed that we used NVIDIA's Founders Edition cards for the RTX 3070 and 3080, but Gigabyte's Turbo model for the 3090. That is because in our qualification testing the Founders Edition cards maxed out at two in a system thanks to venting some of their heat back into the computer chassis. So far the only model we have gotten in which avoids that behavior is the RTX 3090 Turbo from Gigabyte, which uses a single fan in a blower configuration that exhausts almost all of the heat out the back of the computer. That allows us to use up to three of those cards at this time; going to four would draw too much power for a traditional US household circuit.

To test each combination of video cards, we used several different GPU rendering benchmarks. They are listed (with download links) in the specifications table above. For each benchmark, the first chart shows the raw result – while the second, if you click on the right arrow, shows the scaling relative to a single card. Each test was run twice, with the best result being included in the graphs below.

Redshift Demo Results

OctaneBench Results

V-Ray Benchmark Results

Scaling Analysis

As we have seen in the past, GPU rendering scales extremely well across multiple video cards. Both OctaneRender and V-Ray Next have effectively perfect scaling: two cards provide twice the performance of one card, with very slight variations that are within the margin of error. Redshift isn't quite as good, but still very respectable with about a 95% increase in performance per additional card.

The one rendering engine that doesn't do quite so well is the older V-Ray. Scaling from a second video card is still respectable – in the range of 83 to 93% improvement – but adding a third RTX 3090 showed substantially less benefit. Not very many people are likely to be on this older rendering engine, though, and it may be time for us to retire it as a standard part of our benchmark suite. In fact, shortly after we ran these tests Chaos Group put out V-Ray 5 Benchmark, which now also supports RTX technology. We are looking at adding this for the future, but there are some challenges involved with scripting it which will we have to work around before that can happen.

Looking beyond raw performance, let's talk about value. In every one of these benchmarks, a pair of GeForce RTX 3070 cards is able to outperform a single RTX 3090. Since those two 3070s would cost $500 less that seems like an appealing option, and for some folks, it may indeed be viable. There are a couple of things to keep in mind with such comparisons, though:

VRAM – Since the RTX 3070 has only 8GB of onboard memory, compared to the RTX 3090's 24GB, it will be much more limited in terms of scene complexity and texture resolution. Some rendering engines can use system memory for portions of that data, but doing so will usually have a negative impact on performance.
Upgrade Capacity – If you opt for dual RTX 3070s (or even 3080s) then you can't really upgrade further without taking them out and putting in new cards, while a system with a single RTX 3090 blower card could have one or two more added in the future (assuming you planned ahead with the motherboard, chassis size, cooling, and power supply).

It is also worth noting that Gigabyte's RTX 3090 Turbo cards are rather noisy under load, more so than the Founders Edition models. If you are considering a build with them, it would be a good idea to plan on how to minimize the impact of that noise on your workspace.

Conclusion